分布式系统中数据分解的研究

Study of Data Decomposition in Distributed Shared Memory System

数据分解对消息传递并行机下的并行编译器取得高性能至关重要。根据编译器自动得出的数据分解(映射数据到处理机)信息,C语言版本的发送/接收消息循环嵌套可产生出来,从而在处理机之间实现分布数据。不仅一个已被证明且功能强大的数学模型用于产生数据分解代码,而且一个形式化的算法及其实现也已给出。初步实验结果显示该算法能显著提高性能。

Data decomposing is crucial for parallelizing compilers for message-passing parallel machines to achieve high performance. The data decomposition （mapping data onto processors） analysis is automatically from compiler, loop nests in C for arrays that needed to be distributed to send and receive messages between processors are generated, Not only a proven and powerful mathematical model is put into the use of generating code of data decomposition, but also a formal specification of the algorithm as well as its implementation is given. Preliminary experimental results indicate that the algorithm significantly improves the performance.